Connect with us

Business

What is DeepSeek? And How Is It Upending A.I.?

Published

on

What is DeepSeek? And How Is It Upending A.I.?

Tech stocks tumbled. Giant companies like Meta and Nvidia faced a barrage of questions about their future. Tech executives took to social media to proclaim their fears.

And it was all because of a little-known Chinese artificial intelligence start-up called DeepSeek.

DeepSeek caused waves all over the world on Monday as one of its accomplishments — that it had created a very powerful A.I. model with far less money than many A.I. experts thought possible — raised a host of questions, including whether U.S. companies were even competitive in A.I. anymore.

DeepSeek is “AI’s Sputnik moment,” Marc Andreessen, a tech venture capitalist, posted on social media on Sunday.

How could a company that few people had heard of have such an effect?

Advertisement

DeepSeek is a start-up founded and owned by the Chinese stock trading firm High-Flyer. Its goal is to build A.I. technologies along the lines of OpenAI’s ChatGPT chatbot or Google’s Gemini. By 2021, DeepSeek had acquired thousands of computer chips from the U.S. chipmaker Nvidia, which are a fundamental part of any effort to create powerful A.I. systems.

In China, the start-up is known for grabbing young and talented A.I. researchers from top universities, promising high salaries and an opportunity to work on cutting-edge research projects. Both High-Flyer and DeepSeek are run by Liang Wenfeng, a Chinese entrepreneur.

Over the past few years, DeepSeek has released several large language models, which is the kind of technology that underpins chatbots like ChatGPT and Gemini. On Jan. 10, it released its first free chatbot app, which was based on a new model called DeepSeek-V3.

When DeepSeek introduced its DeepSeek-V3 model the day after Christmas, it matched the abilities of the best chatbots from U.S. companies like OpenAI and Google. That alone would have been impressive.

But the team behind the new system also revealed a bigger step forward. In a research paper explaining how it built the technology, DeepSeek said it used only a fraction of the computer chips that leading A.I. companies relied on to train their systems.

Advertisement

The world’s top companies typically train their chatbots with supercomputers that use as many as 16,000 chips or more. DeepSeek’s engineers said they needed only about 2,000 Nvidia chips.

Since late 2022, when OpenAI set off the A.I. boom, the prevailing notion had been that the most powerful A.I. systems could not be built without investing billions of dollars in specialized A.I. chips. That would mean that only the biggest tech companies — such as Microsoft, Google and Meta, all of which are based in the United States — could afford to build the leading technologies.

(The New York Times has sued OpenAI and its partner, Microsoft, claiming copyright infringement of news content related to A.I. systems. The two tech companies have denied the suit’s claims.)

But DeepSeek’s engineers said they needed only about $6 million in raw computing power to train their new system. That was roughly 10 times less than what Meta spent building its latest A.I. technology.

Top A.I. engineers in the United States say that DeepSeek’s research paper laid out clever and impressive ways of building A.I. technology with fewer chips.

Advertisement

In short, the startup’s engineers demonstrated a more efficient way of analyzing data using the chips. Leading A.I. systems learn their skills by pinpointing patterns in huge amounts of data, including text, images and sounds. DeepSeek described a way of spreading this data analysis across several specialized A.I. models — what researchers call a “mixture of experts” method — while minimizing the time lost by moving data from place to place.

Others have used similar methods before, but moving information between the models tended to reduce efficiency. DeepSeek did this in a way that allowed it to use less computing power.

“It has become very clear that other companies, not just someone like OpenAI, can build these kinds of systems,” said Tim Dettmers, a researcher at the Allen Institute for Artificial Intelligence in Seattle and a professor of computer science at Carnegie Mellon University who specializes in building efficient A.I. systems. “DeepSeek used methods that anyone can duplicate.”

DeepSeek’s research paper raised questions about whether big U.S. companies could maintain a significant lead in A.I. Many experts believe that A.I. technology will become a commodity, with many companies selling much the same product.

DeepSeek-V3 can answer questions, solve logic problems and write its own computer programs as effectively as anything already on the market, according to standard benchmark tests.

Advertisement

Just before DeepSeek released its technology, OpenAI had unveiled a new system, called OpenAI o3, which seemed more powerful than DeepSeek-V3. But OpenAI has not released this system to the wider public.

OpenAI o3 was designed to “reason” through problems involving math, science and computer programming. Many experts pointed out that DeepSeek had not built a reasoning model along these lines, which is seen as the future of A.I.

Then on Jan. 20, DeepSeek released its own reasoning model called DeepSeek R1, and it, too, impressed the experts. That eventually sent U.S. investors and others into a panic late last week and over the weekend as they realized the importance of DeepSeek’s new technology.

Yes, it still matters.

Large numbers of A.I. chips can still help companies in many ways. With more chips, they can run more experiments as they explore new ways of building A.I. In other words, more chips can still give companies a technical and competitive advantage.

Advertisement

More chips will also be needed to operate the new breed of “reasoning” A.I. models, experts said. These require more computing power when people and businesses use them.

Yes. To maintain the U.S. lead in the global A.I. race, the Biden administration had put in place rules limiting the number of powerful chips that could be sold to China and other rivals.

But the impressive performance of the DeepSeek model raised questions about the unintended consequences of the American government’s trade restrictions. The controls have forced researchers in China to get creative with a wide range of tools that are freely available on the internet.

Some experts continue to argue in favor of U.S. trade restrictions, saying that they were only recently put in place and that they will have a greater effect on China’s abilities to create A.I. as the years pass.

No. The world has not yet seen OpenAI’s o3 model, and its performance on standard benchmark tests was more impressive than anything else on the market. But experts are concerned that China is jumping ahead on open-source A.I. systems.

Advertisement

Like many other companies, DeepSeek has “open sourced” its latest A.I. system, which means that it has shared the underlying computer code with other businesses and researchers. This allows others to build and distribute their own products using the same technologies.

This is part of the reason DeepSeek and others in China have been able to build competitive A.I. systems so quickly and inexpensively.

In the A.I. world, open source first gathered steam in 2023 when Meta freely shared an A.I. system called Llama. At the time, many assumed that the open-source ecosystem would flourish only if companies like Meta — giant firms with huge data centers filled with specialized chips — continued to open source their technologies.

But DeepSeek and others have shown that this ecosystem can thrive in ways that extend beyond the American tech giants.

Many experts have argued that the big U.S. companies should not open source their technologies because they could be used to spread disinformation or cause other serious harm. Some U.S. lawmakers have explored the possibility of preventing or throttling the practice.

Advertisement

But other experts have argued that if regulators stifle the progress of open-source technology in the United States, China will gain a significant edge. If the best open-source technologies come from China, these experts argue, U.S. researchers and companies will build their systems atop those technologies.

In the long run, that could put China at the heart of A.I. research and development, which could further accelerate its effort to build a wide range of A.I. technologies, including autonomous weapons and other military systems.

Business

Uber, California lawyers say deal reached to avert dueling ballot initiative showdown

Published

on

Uber, California lawyers say deal reached to avert dueling ballot initiative showdown

The state’s trial attorneys and Uber say they have reached a last-minute deal to scrap their dueling ballot measures and avert what was gearing up to be one of most expensive battles of the November election.

The deal, which comes a day after both measures qualified for the November ballot, has Uber agreeing to bulk up safety measures, while the trial attorneys will limit how much they can claim for lien-based medical treatment of victims who get in Uber or Lyft accidents, according to spokespeople for both sides of the campaign.

“Both sides agree: Californians deserve a system that’s safe, fair, and accountable,” read a joint statement from Uber and the Consumer Attorneys of California, a powerful attorney trade group. “This agreement protects patients from unnecessary treatment or getting overcharged, ensures access to medical care and legal representation, and strengthens safety measures.”

The agreement, finalized Thursday, means the ride-share giant will kill its ballot measure to cap how much attorneys can earn in vehicle collision cases and limit medical damages to rates based on insurance. Uber has argued that the costs for medical treatment done on a lien, which allows doctors to get paid from a cut of the plaintiff’s payout, far exceed what it would cost if the victim had used their own insurance.

In return, the Consumer Attorneys of California will cancel its competing ballot measure that sought to increase legal liability for ride-share companies if a passenger is sexually assaulted by a driver. The measure followed an investigation by the New York Times into sexual assault by drivers.

Advertisement

Both sides had poured tens of millions into the campaigns, plastering billboards across Los Angeles.

Lawyers claimed the fight had turned existential with the measure threatening to decimate the profit margin of many personal injury cases and leave drivers with small or thorny cases unable to find an attorney willing to take their case.

Spokespeople say the deal is predicated on their agreement being codified into a bill within the next week. Otherwise, they said, each side will move forward with its ballot measure.

Advertisement
Continue Reading

Business

Commentary: A porn firm that a judge called a ‘copyright troll’ now has Meta in its sights — and it could win

Published

on

Commentary: A porn firm that a judge called a ‘copyright troll’ now has Meta in its sights — and it could win

This porn company made millions by shaming the little guys who downloaded its films. But now it’s going after Meta for copyright infringement.

It isn’t often that a lawsuit can make me smile, much less laugh out loud. The latest exception is Strike 3 Holdings vs. Meta Platforms, which is currently unfolding in San Jose federal court.

Two things are amusing about the case. One is that Meta, the giant social media company, is accused of copyright infringement for allegedly downloading 2,400 of the plaintiff’s movies to train its AI bots. If Meta loses, that would be a serious (and in my opinion, deserved) blow against AI companies that have used copyrighted materials without permission.

The second part of the joke is the identity of the plaintiff. Strike 3 Holdings, you see, makes porn. Moreover, for years it has pursued a plainly unscrupulous business model in which it sues individuals for allegedly downloading its movies without permission, and shames them into settling for a few thousand dollars at a pop.

While it is possible one or more Meta employees downloaded Plaintiffs’ videos, it is just as possible…that a ‘guest, or freeloader,’ or contractor, or vendor, or repair person—or any combination of such persons—was responsible for that activity.

— Meta points the finger at others for a porn scandal

Advertisement

Whether or not Strike 3 has a legitimate claim for copyright infringement, it doesn’t deserve your sympathy. The firm was flayed in 2018 by federal Judge Royce C. Lamberth of Washington, D.C., for engaging in what he labeled a “high-tech shakedown … smacking of extortion.” Lamberth called Strike 3 a “copyright troll” and threw out its lawsuit against an unidentified internet user for having treated his court “not as a citadel of justice, but as an ATM.”

When I wrote about this scheme in 2023, I counted more than 12,440 lawsuits that the Los Angeles-based firm had filed in federal courts coast-to-coast. The latest count, according to a Lexis search a defense lawyer ran for me, is more than 21,000. The vast majority were settled and closed within a few months of their filing, an indication that they were never meant to go to trial.

Now Strike 3 appears to have hooked a big fish. In the first significant ruling in its lawsuit against Meta, the firm scored a surprise win: On June 11, federal Judge Eumi K. Lee of San Jose denied Meta’s motion to dismiss the case. Meta’s defense, she wrote, “strains credulity.”

Advertisement

More about that in a moment. First, a few words about the litigants. Meta needs no introduction: Formerly known as Facebook and based in Menlo Park, Calif., Meta recorded a profit of $60.5 billion last year on $201 billion in revenue.

Strike 3 portrays itself as an avatar of “Hollywood style and quality” in its adult films, which it distributes through its streaming websites such as Blacked, Tushy, Vixen and Wifey. It has described Greg Landry, its former owner and house auteur, as the porn industry’s “answer to Steven Spielberg.”

Neither Meta nor Strike 3 responded to my request for comment beyond the claims and defenses in court filings.

As I reported in 2023, Strike 3 has flooded federal courts with cookie-cutter lawsuits alleging that defendants infringed its copyrights by downloading its movies via BitTorrent, an online service on which unauthorized content can be accessed by almost anyone with an internet connection. Its targets generally have been individuals with plenty to lose from being publicly outed as porn viewers.

“Given the nature of the films at issue,” a federal judge in Connecticut observed last year, “defendants may feel coerced to settle these suits merely to prevent public disclosure of their identifying information, even if they believe they have been misidentified.”

Advertisement

Strike 3’s letters to its target defendants have warned that the statutory penalty for willful copyright infringement is $150,000, but offer to make the case go quietly away for a few thousand bucks, which would be a fraction of the cost of hiring a defense lawyer, not to mention the downside of exposing oneself as a porn fiend.

J. Curtis Edmondson, a Portland, Ore., lawyer who won a case against Strike 3, estimated in 2023 that Strike 3 “pulls in about $15 million to $20 million a year from its lawsuits.” But financial data that could validate his estimate hasn’t surfaced in court records.

There’s nothing new about content owners’ aggressive pursuit of copyright infringers. The practice was pioneered by the Recording Industry Assn. of America, when the industry feared that unauthorized downloading of music through programs such as Napster threatened its very existence. From 2003 through 2008, the association sued some 35,000 alleged song pirates.

But it abandoned the strategy because its legal dragnet swept up sympathetic targets such as single mothers and teenage girls, creating a public relations disaster.

There followed the appearance of outright trolls such as Prenda Law Group, which posted porn films online as bait to attract downloaders, whom it then sued in what judges ultimately found to be sham lawsuits. Prenda principal John L. Steele even bragged publicly that Prenda had made nearly $15 million with its lawsuits. U.S. Judge Otis Wright II of Los Angeles put the kibosh to its practice by slapping the Prenda lawyers with stiff sanctions for contempt.

Advertisement

That brings us to Strike 3’s case against Meta, which it filed in July. Strike 3 hasn’t been accused of a Prenda-style fraud, since it does own the films at issue and its right to sue copyright infringers isn’t disputed. But its allegation that Meta downloaded its films to train its AI bots, rather than just for personal enjoyment, is a new wrinkle for an old issue.

Strike 3 says its lawsuit grew out of a separate case in which a witness testified that Meta had downloaded thousands of pirated books to train its LLaMA AI bots — that is, feeding the content into LLaMA for it to use to generate answers to user questions. (Numerous lawsuits have been filed against AI firms alleging similar infringement.)

Strike 3 says that case prompted it to look into whether Meta had downloaded any of its content. It says it discovered that 47 IP addresses owned by Meta — that is, digital identifiers of internet accounts — had downloaded its movies without permission.

In all, Strike 3 alleges, those Meta addresses downloaded at least 2,396 of its movies — almost its entire catalog — more than 6,000 times via BitTorrent. What’s more, Strike 3 says Meta then posted some of that content back onto BitTorrent to take advantage of BitTorrent’s “tit-for-tat” mechanism through which users can obtain faster download speeds by uploading content to the platform.

If Strike 3 were to prevail on all its claims for illicit downloading, it would be entitled to about $360 million in damages, observes Eric Fruits, an Oregon economist who has testified for the defense in some Strike 3 lawsuits.

Advertisement

One might ask why Meta might be downloading porn for any reason, bot-training or otherwise. Meta, in its defense filings, says Strike 3 has offered no proof that Meta, as a corporation, was responsible for the downloading. If it happened, Meta says, it would have been inadvertent.

“Tens of thousands of employees and innumerable contractors, visitors, and third parties access the internet at Meta every day,” it wrote in its motion to dismiss the case. “While it is possible one or more Meta employees downloaded Plaintiffs’ videos, it is just as possible … that a ‘guest, or freeloader,’ or contractor, or vendor, or repair person — or any combination of such persons — was responsible for that activity.” The “sporadic downloads,” Meta says, “exhibit the hallmarks of personal use,” not corporate strategy.

This defense has borne fruit in other Strike 3 cases, in which defendants successfully argued simply having an IP address that was used to infringe wasn’t enough to prove they committed the infringements.

Strike 3 says it can show that the downloads weren’t the work of random users. Some downloads, it says, were coordinated among several Meta IP addresses, all based on the same algorithmic keywords and occurring simultaneously, suggesting that the infringements “took place within Meta’s walls.”

On Dec. 15, 2022, for instance, downloads apparently based on the keyword “teen” involved not only the movies “Teenage Mutant Ninja Turtles” and “Teen Titans Go to the Movies,” but also “Teen Sex Sessions 2” and “Teens love Tats XXX,” according to Lee’s ruling. Other simultaneous downloads swept up episodes of “The Big Bang Theory” and “Ted Lasso” out of order, though a putative human user would probably have downloaded them sequentially.

Advertisement

“It strains credulity,” Lee ruled, “to suggest that these correlations are mere coincidence and the product of individual human selections.” Rather, the use of an algorithm would account for “why pornography was downloaded alongside children’s cartoons and sitcoms. … The odds that multiple people using the Corporate IP addresses … coincidentally torrented the same show, rather than simply streaming it, on the exact same day strains belief.”

The case is still at an early stage. For Strike 3, the lawsuit offers the potential of a big score. But Meta has signaled that it’s not inclined to roll over like a family man caught downloading skin flicks and worrying about his reputation at home and around town.

This time, Strike 3 may have a fight on its hands with a defendant that has money to burn.

Advertisement
Continue Reading

Business

Rivian lays off hundreds of workers days after new vehicle deliveries begin

Published

on

Rivian lays off hundreds of workers days after new vehicle deliveries begin

Rivian said it’s laying off hundreds of employees, or less than 2% of its workforce, as part of restructuring efforts aimed at making the company profitable for the first time.

The layoffs come one week after the Irvine-based electric vehicle maker began deliveries of its highly anticipated R2 SUV.

The company is hoping that the R2, which is currently only available as a performance version for $57,990, could attract more customers with its lower price tag.

But industry analysts said the performance R2 is still not affordable for many Americans, and investors reacted with disappointment to the first deliveries June 9, with shares falling 7% that day. On Wednesday, Rivian shares gained .33 points, or 2%, to close at $16.26.

The company said a standard version of the R2 starting at $44,990 will become available next year.

Advertisement

The layoffs took effect on Monday and affected Rivian’s service and customer organization employees, including sales and marketing teams. Rivian employed 15,232 people as of December.

“We recently restructured a handful of teams within Rivian as we work to profitably scale our ‌business,” a ⁠company spokesperson said.

The laid off employees have been provided with severance packages and are encouraged to apply for other open roles with Rivian, the company said.

Rivian may be trying to reach profitability by saving money on labor, said Ivan Drury, director of insights at Edmunds.

“You have to wonder to what degree they do plan on replacing those people with some level of AI and automation,” he said.

Advertisement

Rivian, which is pouring money into autonomous vehicle efforts including a robotaxi partnership with Uber, has struggled to turn a profit with its luxury EVs.

The layoffs are likely not directly tied to recent reception of the R2, auto analyst Brian Moody said.

“I think that it’s declining interest in new electric cars, and maybe declining interest in expensive things,” he said. “We can surmise that [layoff] process began long before the R2 launch.”

The company lost $3.6 billion last year and recently said it is no longer expecting to meet its 2027 adjusted core profit target.

There has been a broad cooling of the EV market. Major automakers including Honda and Ford have cut back their EV options as excitement for the vehicles has fallen under the Trump administration. A $7,500 EV tax credit for new vehicles expired in September.

Advertisement

Rivian cut 4.5% of its workforce in October, or more than 600 jobs, following the expiration of the credit. The company also laid off about 200 employees in September.

In a recent turnaround, Rivian surprised the market with strong earnings results in February, reporting gross profits for 2025 of $144 million compared with a net loss in 2024 of $1.2 billion. Gross profit is revenue without subtracting the cost of production expenses.

In its earnings release, Rivian credited the swing to “strong software and services performance, higher average selling prices, and reductions in cost per vehicle.”

“The company has never posted a full year’s worth of profit, and this is the one lever they can pull to rightsize things,” Drury said.

Advertisement
Continue Reading
Advertisement

Trending